CSIRO RTI Series (6): Image Reconstruction

Image Reconstruction

I followed the core reconstruction concept from:

Wilson & Patwari, Radio tomographic imaging with wireless networks [1]

The model links RSS-based measurements to an attenuation image:

Where:

• y: vector of RSS measurement differences
• W: weighting matrix
• x: attenuation image to estimate (in dB)
• Cₓ: prior covariance matrix
• σₙ⁻²: node variance term

I skip the full derivation (well covered in the paper) and focus on how I defined y in practice.

Three y-vector strategies

1) Cycle-to-cycle difference: (T_n - T_{n-1})

• Best spatial accuracy during continuous movement
• Weakness: when an object stops moving, it can fade from the image as the network adapts

2) Baseline difference: (T_n - T_0)

• Better for detecting slow or static objects
• Weakness: baseline drifts with environmental changes, so long runs become less reliable

3) Standard deviation-based y

• Similar quality to method 1
• Improved as iteration count increased

Conclusion

I used method 1 as the default because it gave the most stable overall results in my setup. Methods 2 and 3 are still useful alternatives depending on environment stability and runtime conditions.

---

Reference

[1] Wilson, Joey, and Neal Patwari. “Radio tomographic imaging with wireless networks.” IEEE Transactions on Mobile Computing 9.5 (2010): 621–632.

2026 Update Note

• Migrated and language-polished in 2026.
• The three y-vector strategies remain useful framing choices for RSSI-based reconstruction trade-offs.
• In current practice, this stage can be further improved with regularization tuning and adaptive baseline management.